OpenCV Image Processing - Getting Started

In [1]:
import cv2
import numpy as np
from matplotlib import pyplot as plt

%matplotlib inline
In [2]:
plt.rcParams["figure.figsize"] = [10.0,10.0]

Loading Image

Loading an image with OpenCV and displaying it with MatPlotLib. We can access the images attributes as well.

In [3]:
img = cv2.imread("opencv-img-processing/img.pgm")

print(img[50,50])
print(img.shape)
print(img.size)
print(img.dtype)

[130 130 130]
(360, 460, 3)
496800
uint8

We can display the image inline in this notebook using matplotlib.

In [4]:
plt.title('Casablanca')
plt.imshow(img)
Out[4]:
<matplotlib.image.AxesImage at 0x7fe5a8709630>

Image Region Of Interest (ROI)

We can get a section of an image using NumPy notation

In [5]:
width,height,_ = img.shape

img_partial = img[int(width*1/4):int(width*3/4),int(height*1/4):int(height*3/4)]
plt.title('Casablanca Partial')
plt.imshow(img_partial)
Out[5]:
<matplotlib.image.AxesImage at 0x7fe5a86e0c50>

Colored Images

We can also read colored images, but be aware that the color scheme used by default by OpenCV is BGR (as opposed to the more widely known RGB).

In [6]:
img_color = cv2.imread("opencv-img-processing/color.tiff")
plt.title('Peppers Inverted Color Scheme (BGR)')
plt.imshow(img_color)
Out[6]:
<matplotlib.image.AxesImage at 0x7fe5a864a780>

Despite that, conversion can be done in OpenCV.

In [7]:
img_color = cv2.cvtColor(img_color,cv2.COLOR_BGR2RGB)
plt.title('Peppers')
plt.imshow(img_color)
Out[7]:
<matplotlib.image.AxesImage at 0x7fe5a85ce908>

Color Channels

Obtainning separate color channels from an image is simple. It can be done by using OpenCV methods;

In [8]:
r, g, b= cv2.split(img_color)
b.fill(0)
g.fill(0)

plt.title('Red Peppers')
plt.imshow(cv2.merge((r, g, b)))
Out[8]:
<matplotlib.image.AxesImage at 0x7fe5a85a7d68>

Or even by using Numpy notation.

In [9]:
plt.title('Green Peppers')

plt.imshow(img_color[:,:,1])
Out[9]:
<matplotlib.image.AxesImage at 0x7fe5a402bac8>

Editing image channels can also be done via Numpy notation.

In [10]:
img_color[:,:,0] = 0
img_color[:,:,1] = 0
plt.title('Blue Peppers')
plt.imshow(img_color)
Out[10]:
<matplotlib.image.AxesImage at 0x7fe5992dddd8>

Image Operations

OpenCV provides methods for simple image operations.

In the example below we add two images.

In [11]:
img_color = cv2.imread("opencv-img-processing/color.tiff")
img_color = cv2.cvtColor(img_color,cv2.COLOR_BGR2RGB)

img_ruler = cv2.imread("opencv-img-processing/ruler.tiff")
img_ruler = cv2.cvtColor(img_ruler,cv2.COLOR_BGR2RGB)

plt.title('Ruler')
plt.imshow(img_ruler)
Out[11]:
<matplotlib.image.AxesImage at 0x7fe5992bd198>
In [12]:
img_merged = cv2.addWeighted(img_color,0.5,img_ruler,0.5,0)
plt.title('Peppers Rule')
plt.imshow(img_merged)
Out[12]:
<matplotlib.image.AxesImage at 0x7fe5903cde48>

Image Bitwise Operations

OpenCV also provides bitwise operations across all image channels, making it possible (and simple) to do processes such as obtaining a mask and operating over that mask.

In [13]:
img_saturn = cv2.imread("opencv-img-processing/saturn.tiff")
img_saturn = cv2.cvtColor(img_saturn,cv2.COLOR_BGR2RGB)
plt.title('Satturn')
plt.imshow(img_saturn, aspect='equal')
Out[13]:
<matplotlib.image.AxesImage at 0x7fe5903b0198>
In [14]:
img_saturn_gray = cv2.cvtColor(img_saturn, cv2.COLOR_BGR2GRAY)
ret, mask = cv2.threshold(img_saturn_gray, 20,255,cv2.THRESH_BINARY_INV)
mask_inv = cv2.bitwise_not(mask)
plt.title('Satturn Mask')
plt.imshow(mask_inv)
plt.colorbar()
Out[14]:
<matplotlib.colorbar.Colorbar at 0x7fe59032cf60>
In [15]:
row,col,_ = img_saturn.shape
color_roi = img_color[0:row,0:col]

img_masked = cv2.bitwise_and(color_roi, color_roi, mask = mask_inv)
plt.title('Pepper Satturn Masked')
plt.imshow(img_masked)
Out[15]:
<matplotlib.image.AxesImage at 0x7fe5902a5ba8>
In [16]:
img_added = cv2.add(img_saturn,img_masked)
plt.title('Pepper Satturn Added')
plt.imshow(img_added)
Out[16]:
<matplotlib.image.AxesImage at 0x7fe590202860>
In [17]:
img_added_weighted = cv2.addWeighted(img_saturn,0.5,img_masked,0.5,0)
plt.title('Pepper Satturn Added Weighted')
plt.imshow(img_added_weighted)
Out[17]:
<matplotlib.image.AxesImage at 0x7fe5901d9be0>

The snippets in this notebook can be seen in a python script with the same name located at the opencv-img-processing folder. The script will not be using matplotlib in favor of OpenCV.